behavioural

Behavioural Patterns

Source: Gang of Four — Design Patterns (1994)

Pain Signals — You Are in This Branch When:

• A method keeps growing if/switch branches for each new case or plan
• The same algorithm is implemented differently in multiple places
• A pipeline of steps needs to be reordered, skipped, or extended without touching the core
• An object behaves completely differently depending on what state it is in
• Multiple objects need to react when one changes, but coupling them directly is fragile
• Actions need to be deferred, logged, retried, or undone
• You need to add new operations to a stable structure without modifying it

---

Chain of Responsibility

Passes a request through a sequence of handlers. Each handler decides to process it, stop it, or pass it forward.

Pain it removes: a single method accumulating all validation and processing steps; steps that need to be reordered or inserted without changing the core.

When to use: middleware pipelines, request validation sequences, approval workflows where each level may approve or escalate.

class Handler
  def initialize(next_handler = nil)
    @next = next_handler
  end

  def call(request)
    return unless handle?(request)
    @next&.call(request)
  end
end

class RateLimitHandler < Handler
  def handle?(request)
    return true if within_limit?(request)
    request.halt(429, "Too Many Requests")
    false
  end
end

class AuthHandler < Handler
  def handle?(request)
    return true if authenticated?(request)
    request.halt(401, "Unauthorized")
    false
  end
end

# Compose the chain
pipeline = RateLimitHandler.new(AuthHandler.new(RequestHandler.new))
pipeline.call(request)

When NOT to use: when steps always execute in sequence with no possibility of stopping (use a plain sequence of method calls instead).

Health check: each handler has one responsibility and a clear stop/continue contract. Breaks down when handlers mutate shared state or depend on each other's internals.

---

Command

Encapsulates a request as an object, decoupling the sender from the receiver and enabling queuing, logging, retry, and undo.

Pain it removes: actions that need to be deferred, retried, audited, or undone cannot be done with direct method calls.

When to use: job queues, undo/redo stacks, audit logs, "run later" workflows, transactional outbox pattern.

class SaveDocumentCommand:
    def __init__(self, document_id: str, content: str):
        self.document_id = document_id
        self.content = content

    def execute(self, service: DocumentService) -> None:
        service.save(self.document_id, self.content)

    def undo(self, service: DocumentService) -> None:
        service.restore_previous(self.document_id)

# Commands are objects — can be queued, serialized, retried
queue.push(SaveDocumentCommand(doc_id, content))

When NOT to use: when the operation is simple and immediate with no need for queuing, retry, or undo.

Distinction from Strategy: Command is an action (something that happened or will happen). Strategy is an algorithm (a way of doing something).

---

Strategy

Defines a family of interchangeable algorithms behind a common interface. The caller selects the algorithm; switching algorithms does not change the caller.

Pain it removes: if plan == 'X': do_this() elif plan == 'Y': do_that() — branching that grows with each new variant and duplicates test setup.

When to use: payment providers, export formats, routing algorithms, rate limiting policies, notification channels, pricing rules per customer tier.

class NotificationChannel:
    def send(self, user: User, message: str) -> None:
        raise NotImplementedError

class EmailChannel(NotificationChannel):
    def send(self, user: User, message: str) -> None:
        self._smtp.send(user.email, message)

class SmsChannel(NotificationChannel):
    def send(self, user: User, message: str) -> None:
        self._twilio.send(user.phone, message)

class NotificationService:
    def __init__(self, channel: NotificationChannel):
        self._channel = channel  # injected — caller stays stable

    def notify(self, user: User, message: str) -> None:
        self._channel.send(user, message)

When NOT to use: when there is only one algorithm now and no real expectation of variation. Premature Strategy adds indirection for no gain.

Signal it's working: adding a new variant requires adding one class and updating the selector — nothing else changes.

---

State

Allows an object to alter its behaviour when its internal state changes. The object appears to change its class.

Pain it removes: methods full of if self.status == 'draft': ... elif self.status == 'submitted': ... that multiply with every new state.

When to use: objects with well-defined modes and transitions — order lifecycle, connection states, approval workflows, subscription status, document editing modes.

class Order:
    def __init__(self):
        self._state = DraftState(self)

    def submit(self):
        self._state.submit()

    def cancel(self):
        self._state.cancel()

    def transition_to(self, state: OrderState):
        self._state = state

class DraftState:
    def submit(self):
        self.order.transition_to(SubmittedState(self.order))

    def cancel(self):
        self.order.transition_to(CancelledState(self.order))

class SubmittedState:
    def submit(self):
        raise InvalidTransition("Already submitted")

    def cancel(self):
        self.order.transition_to(CancelledState(self.order))

When NOT to use: when transitions are simple and few — a field and a conditional is clearer than a full State implementation.

Distinction from Strategy: Strategy is selected by the caller and held stable. State transitions itself based on internal rules.

---

Observer

Defines a one-to-many dependency so that when one object changes state, all dependents are notified automatically.

Pain it removes: the subject needing to know about and directly call every interested party — tight coupling that makes adding subscribers painful.

When to use: event-driven UI updates, domain event notification, logging/metrics hooks, cache invalidation on data change.

class EventBus:
    def __init__(self):
        self._subscribers: dict[str, list[Callable]] = {}

    def subscribe(self, event_type: str, handler: Callable) -> None:
        self._subscribers.setdefault(event_type, []).append(handler)

    def publish(self, event_type: str, event: Any) -> None:
        for handler in self._subscribers.get(event_type, []):
            handler(event)

# Subject publishes; observers subscribe — no direct coupling
bus.subscribe('order.submitted', send_confirmation_email)
bus.subscribe('order.submitted', reserve_inventory)
bus.subscribe('order.submitted', update_analytics)

When NOT to use: when the reaction is always the same single operation (just call it directly). Observer hides control flow — keep subscribers visible and documented.

Risk: surprising side effects from hidden observers. Every subscriber must be documented; avoid observers that trigger further observers.

---

Memento

Captures and externalises an object's internal state so it can be restored later, without violating encapsulation.

Pain it removes: implementing undo/rollback requires exposing internal state or duplicating save logic across callers.

When to use: undo/redo stacks, draft-vs-published versioning, rollback on failure, wizard-style multi-step forms.

@dataclass(frozen=True)
class EditorMemento:
    content: str
    cursor_position: int
    # immutable snapshot — internals not exposed as mutable

class Editor:
    def save(self) -> EditorMemento:
        return EditorMemento(self._content, self._cursor)

    def restore(self, memento: EditorMemento) -> None:
        self._content = memento.content
        self._cursor = memento.cursor_position

class History:
    def __init__(self):
        self._stack: list[EditorMemento] = []

    def push(self, memento: EditorMemento) -> None:
        self._stack.append(memento)

    def pop(self) -> EditorMemento:
        return self._stack.pop()

When NOT to use: when state is large and snapshots are expensive — consider event sourcing (record deltas, not full snapshots) instead.

---

Mediator

Defines an object that encapsulates how a set of objects interact, promoting loose coupling by preventing direct references between them.

Pain it removes: objects that know too much about each other — changing one requires changing several others because they call each other directly.

When to use: complex UI coordination (form fields affecting each other's state), workflow orchestration across multiple components, chat systems, air traffic control-style coordination.

class FormMediator:
    def __init__(self, country_field, region_field, postal_field):
        self._country = country_field
        self._region = region_field
        self._postal = postal_field

    def on_country_changed(self, country: str) -> None:
        self._region.load_options_for(country)
        self._postal.set_format_for(country)
        self._region.clear()
        self._postal.clear()

When NOT to use: when the coordination logic is simple. Mediator risks becoming a "god object" — keep its responsibilities narrow and well-defined.

---

Visitor

Lets you add new operations to an object structure without modifying the classes in that structure.

Pain it removes: needing to add new operations to a stable hierarchy without modifying every class in it — the classic problem with ASTs, document structures, and expression trees.

When to use: the object structure is stable (rarely gains new types) but operations on it change frequently; compiler passes, document export, tax calculation over an order structure.

class OrderVisitor:
    def visit_line_item(self, item: LineItem) -> None: ...
    def visit_discount(self, discount: Discount) -> None: ...
    def visit_shipping(self, shipping: Shipping) -> None: ...

class TaxCalculatorVisitor(OrderVisitor):
    def __init__(self):
        self.total_tax = Money.zero()

    def visit_line_item(self, item: LineItem) -> None:
        self.total_tax += item.price * item.tax_rate

    def visit_discount(self, discount: Discount) -> None:
        pass  # discounts don't affect tax base in this jurisdiction

class OrderElement:
    def accept(self, visitor: OrderVisitor) -> None:
        raise NotImplementedError

When NOT to use: when the structure changes frequently (adding new element types requires updating every visitor). In application code, Strategy or Chain of Responsibility are usually simpler.

---

Iterator

Provides a way to sequentially access elements of a collection without exposing its underlying representation.

Pain it removes: callers needing to know the internal structure of a collection to traverse it; multiple traversal strategies for the same structure.

When to use: custom collections with non-trivial traversal (trees, graphs, paginated remote results, lazy-loaded sequences).

class PaginatedResultIterator:
    def __init__(self, fetch_page: Callable, page_size: int = 50):
        self._fetch = fetch_page
        self._page_size = page_size
        self._page = 0
        self._buffer = []
        self._exhausted = False

    def __iter__(self):
        return self

    def __next__(self):
        if not self._buffer:
            if self._exhausted:
                raise StopIteration
            results = self._fetch(page=self._page, size=self._page_size)
            self._page += 1
            if len(results) < self._page_size:
                self._exhausted = True
            self._buffer = results
        if not self._buffer:
            raise StopIteration
        return self._buffer.pop(0)

When NOT to use: when the language's built-in iteration primitives (list, generator, stream) already solve the problem. Only build a custom iterator when the traversal logic is genuinely non-trivial.

---

Template Method

Defines the skeleton of an algorithm in a base class, deferring some steps to subclasses. Subclasses can override specific steps without changing the overall structure.

Pain it removes: the same algorithm structure repeated in multiple subclasses with only a few steps varying — duplication that diverges over time.

When to use: report generation pipelines, data import/export workflows, test setup/teardown patterns, HTTP request lifecycle hooks.

class DataImporter:
    def run(self) -> ImportResult:
        """Template method — skeleton is fixed, steps are overridden."""
        raw = self.fetch()
        validated = self.validate(raw)
        transformed = self.transform(validated)
        return self.persist(transformed)

    def fetch(self) -> RawData:
        raise NotImplementedError

    def validate(self, raw: RawData) -> ValidatedData:
        # Default: pass through — subclasses override if needed
        return ValidatedData(raw)

    def transform(self, data: ValidatedData) -> DomainObjects:
        raise NotImplementedError

    def persist(self, objects: DomainObjects) -> ImportResult:
        raise NotImplementedError

class CsvOrderImporter(DataImporter):
    def fetch(self) -> RawData:
        return self._csv_reader.read(self._file_path)

    def transform(self, data: ValidatedData) -> DomainObjects:
        return [Order.from_csv_row(row) for row in data.rows]

When NOT to use: when the variation is better expressed as a Strategy (composition) than subclassing (inheritance). Prefer Strategy when you need to swap behaviour at runtime; use Template Method when the structure is fixed and variations are compile-time.

Distinction from Strategy: Template Method uses inheritance — the skeleton and steps live in a class hierarchy. Strategy uses composition — the algorithm is injected.

---

Design Checklist

• Pattern chosen in response to a named, existing pain — not speculatively
• Chain of Responsibility: each handler has one responsibility and a clear stop/continue contract
• Command: actions are serialisable or at minimum recoverable on failure
• Strategy: adding a new variant touches only one new class and the selector
• State: all valid transitions are documented; invalid transitions raise explicitly
• Observer: all subscribers are documented; side effects are visible
• Memento: snapshots are immutable; restore does not expose internals
• Mediator: coordinator has a narrow, named responsibility — not a catch-all
• Visitor: used only when structure is stable and operations vary
• Template Method preferred only when variation is compile-time; Strategy preferred for runtime